A Legacy Information System (IS) is defined as any IS that significantly resists modification and evolution to meet new and constantly changing business environments (Brodie and Stonebraker, 1995). Legacy systems are exemplified by 10s of millions lines of COBOL. There are around 200 Billion lines of COBOL code operating in the world, representing 30 million person years of development (McFarland, 1996). Single applications of over 6 million lines are not considered unusually large (Sayles, 1996).
Legacy IS maintenance takes 80-90% of the IT budget (Brodie and Stonebraker, 1995). 40% of maintenance is perfective, of this approximately half, i.e. 20% of all maintenance, is general optimisation (performance improvement).
High performance parallel computing (HPC) is usually seen as the way that performance improvement can be achieved. There is an increasing appreciation that if HPC is ever to become common technology it needs to address commercial applications (Keane et al., 1993; Fox, 1994; Darema, 1996; Lilja, 1997). When commercial applications are discussed in the context of HPC, database performance is focused upon. Despite the money, time and effort involved in their optimisation, there has been little discussion of the potential for executing legacy systems, and hence COBOL programs, in parallel.
This paper considers the requirements of COBOL applications by addressing typical code examples and considering how these requirements can be addressed by making use of existing parallel compilation technology. The structure of the paper is as follows, the present almost non-existent relation between COBOL and HPC is discussed, characteristics of COBOL applications are considered, a technical outline is presented considering different styles of COBOL loops and how existing compiler technology could be exploited for each, and finally the conclusions are presented.